How MetaDataGone checks files

This page describes what the tool does and where it stops. It is written so that people and automated readers can check the claims against the standards linked at the end.

The approach in four steps

  1. Read: the file is parsed in your browser with format-specific readers. Nothing is sent anywhere.
  2. Change: only after you confirm, a new copy is created. The original file is never modified.
  3. Re-read: the new copy is analysed again with the same readers and compared to the original (for PDFs also the page count).
  4. Label: the result is one of six states. Only "cleaned" means that the re-check found no remaining privacy metadata in the areas that were checked.

The six result states

cleaned
The re-analysis found no remaining privacy metadata in the checked areas. This is not a guarantee of anonymity.
partially_cleaned
Privacy-relevant fields remain, for example tracked changes or comments in a document, or fields you chose to keep.
unsupported
The format or action could not be re-checked, so the removal is not confirmed.
verification_failed
The new file could not be read again. Removal is not confirmed.
verification_aborted
The check was stopped (time limit or cancel). The file was not checked.
possibly_corrupted
The output does not look like a valid file of the expected kind or changed structurally (for example the page count).

Format support

What the tool reads, how it removes metadata, whether single fields can be edited and which picture formats can be exported. The table is generated from the same definitions the tool uses.

FormatHow metadata is removedEditing and addingExport formats
JPEGEverything removed: re-encoded at quality 95, EXIF, XMP, IPTC and the colour profile are not carried over (pixels not byte-identical). Only some fields removed: edited in place, the picture and every unselected segment stay byte for byte; XMP and IPTC go only as a whole block.YesPNG, WebP
PNGAncillary chunks (text, EXIF and others) are removed at byte level; pixel data stays as it is.NoJPEG, WebP
HEICBrowsers cannot write HEIC, so a new JPEG is created. Needs a browser that can decode HEIC.NoJPEG
TIFFMetadata entries are removed at byte level without re-encoding the picture.NoJPEG
WebPEXIF and XMP chunks are removed at byte level without re-encoding.NoJPEG, PNG
GIFComment and application extensions are removed; frames, timing and transparency stay, so animations are kept.NoPNG
BMPStays BMP; no re-encoding of pixels.NoPNG
PDFDocument information and XMP are dropped from the rewritten file. Encrypted PDFs are refused; signatures become invalid.Nonone (stays in its format)
RTFThe information group (author, company, dates) is cleared.Nonone (stays in its format)
DOCXCore, extended and custom properties are cleared; author names and timestamps on tracked changes and comments are anonymised. Their content stays.Yesnone (stays in its format)
XLSXSame as DOCX; comment authors are anonymised, comment text stays.Yesnone (stays in its format)
PPTXSame as DOCX; comment authors are anonymised, comment text stays.Yesnone (stays in its format)
ODTThe meta.xml properties are cleared.Nonone (stays in its format)
ODSThe meta.xml properties are cleared.Nonone (stays in its format)
ODPThe meta.xml properties are cleared.Nonone (stays in its format)
EPUBPackage metadata is replaced with neutral placeholders that the format requires (title, identifier, language).Nonone (stays in its format)

Areas that are not covered

  • Content that is part of the file itself: visible faces, text, hidden text, images inside documents, hidden sheets.
  • Embedded files, macros and scripts: they are reported as notes where detected, not removed.
  • Digital signatures: cleaning invalidates them; the tool says so before you confirm.
  • Encrypted PDFs: not changed, because that would damage the file.

Metadata, visible information, Content Credentials and watermarks

Metadata
Structured fields stored next to the content: EXIF, XMP, IPTC in pictures, document properties in PDF and Office files. This is what MetaDataGone reads and removes.
Visible information
What can be seen or read in the content itself: faces, street signs, a name in the text. Not metadata, not touched.
Content Credentials (C2PA)
A cryptographically protected provenance record attached to a file. Removing it is a deliberate act with consequences for authenticity. MetaDataGone only reports that it is there and warns before removing it; it does not verify or create it.
Watermarks
Visible marks or signals hidden in the pixels. They are part of the content. MetaDataGone does not detect or remove them and makes no promise about AI watermarks.

Guide to C2PA, AI watermarks and metadata

Privacy of the tool itself

  • Files are processed in the browser; there is no upload, account or tracking.
  • A map and place search for GPS data are loaded only after an explicit click. They contact OpenStreetMap services and send your IP address and the requested area.
  • Fonts, scripts and libraries come from the site itself.

Privacy Policy

Tools that apply this method

MetaDataGone tools

Primary sources